Meta-Learning Divergences of Variational Inference

TL;DR

This paper introduces a meta-learning approach to automatically select divergence measures for variational inference, improving approximation quality across various tasks including Bayesian neural networks and image generation.

Contribution

It proposes a novel meta-learning algorithm that learns divergence metrics and variational initializations, enhancing VI performance without extra cost.

Findings

Outperforms standard VI in Gaussian mixture approximation

Improves Bayesian neural network regression results

Enhances image generation with variational autoencoders

Abstract

Variational inference (VI) plays an essential role in approximate Bayesian inference due to its computational efficiency and broad applicability. Crucial to the performance of VI is the selection of the associated divergence measure, as VI approximates the intractable distribution by minimizing this divergence. In this paper we propose a meta-learning algorithm to learn the divergence metric suited for the task of interest, automating the design of VI methods. In addition, we learn the initialization of the variational parameters without additional cost when our method is deployed in the few-shot learning scenarios. We demonstrate our approach outperforms standard VI on Gaussian mixture distribution approximation, Bayesian neural network regression, image generation with variational autoencoders and recommender systems with a partial variational autoencoder.